Computational methods, clinical evidence, and translational readiness for artificial intelligence for HER2 assessment using silver in situ hybridization (SISH).

Accurate determination of human epidermal growth factor receptor 2 (HER2) gene amplification is a cornerstone of precision oncology in breast cancer and an expanding range of solid tumors. While immunohistochemistry (IHC) remains the first-line screening modality, in situ hybridization (ISH) techniques are essential for resolving equivocal cases and providing quantitative assessment of HER2 gene status. Among ISH methods, silver-enhanced in situ hybridization (SISH) has emerged as a clinically relevant bright-field alternative to fluorescence in situ hybridization (FISH), offering permanent staining, high concordance, and compatibility with routine histopathology workflows. Despite these advantages, SISH interpretation remains labor-intensive and subject to interobserver variability, particularly in borderline and heterogeneous tumors. The rise of digital pathology and artificial intelligence (AI) has enabled automated analysis of whole-slide images (WSIs), supporting tasks such as nuclei segmentation, signal detection, HER2 copy-number estimation, and HER2/CEP17 ratio computation. However, current evidence remains heterogeneous, with limitations including small datasets, limited external validation, and insufficient methodological transparency. This systematic review synthesizes current literature on AI-assisted HER2 assessment with a specific focus on SISH and related bright-field ISH modalities. We examine computational workflows, validation strategies, performance benchmarks, and translational readiness while critically appraising limitations in the evidence base. The analysis demonstrates that AI can support reproducible and scalable HER2 assessment when aligned with clinical guidelines, but emphasizes that clinical deployment requires rigorous validation, uncertainty-aware design, and integration with expert pathology workflows.